IBM's Watson supercomputer creates a movie trailer

IBM's Watson supercomputer cre...
Anya Taylor-Joy and Kate Mara in Morgan (2016)
Anya Taylor-Joy and Kate Mara in Morgan (2016)
View 2 Images
A graph highlighting the moments of interest Watson selected after processing the entire feature length film.
A graph highlighting the moments of interest Watson selected after processing the entire feature length film.
Anya Taylor-Joy and Kate Mara in Morgan (2016)
Anya Taylor-Joy and Kate Mara in Morgan (2016)

What happens when an artificial intelligence system is tasked with making a trailer to a horror movie about an artificially enhanced human hybrid? We now have the answer.

Morgan, a new suspense/horror film from debut director Luke Scott (son of Ridley Scott), from all reports plays like a mix between last year's impressive Ex Machina and a modern-day riff on Mary Shelley's Frankenstein. Considering the film is about a human hybrid packed with nanotechnology 20th Century Fox decided to approach IBM Research and see if its Watson supercomputer could come up with a machine-generated trailer.

Recently we saw the Impossible Things project attempt to develop an AI that could write a horror movie but the results weren't great and the initial trailer came across as a cliched compendium of generic stereotypes. In this instance the IBM researchers took a more comprehensive approach, training Watson by breaking down 100 horror movie trailers into separate scenes and then having the system analyze each sequence through three different prisms.

A graph highlighting the moments of interest Watson selected after processing the entire feature length film.
A graph highlighting the moments of interest Watson selected after processing the entire feature length film.

First Watson completed a visual analysis of each trailer moment and assigned individual scenes with a variety of moods and labels. Then the audio of each scene was studied, again to gauge the mood and sentiment of the given moment. Finally Watson deconstructed the composition of each scene in order to develop a sense of the locations and specific shots (i.e. lighting and framing) generally utilized in successful horror film trailers.

Once Watson was acquainted with the general strategies involved in compiling a horror trailer the researchers let it "watch" the full-length feature from beginning to end resulting in it selecting 10 key moments that the ideal trailer should consist of. From this point a human editor came on board to construct the final cut from the six minutes of footage selected.

Morgan | IBM Creates First Movie Trailer by AI [HD] | 20th Century FOX

The final result is reasonably interesting, although it's most certainly not as dynamic as a regular human created trailer and I could happily live the rest of my life without hearing another creepy nursery rhyme sung by a child. The most refreshing aspect of Watson's trailer is how little plot it gives away. Over recent years a common criticism lobbed at modern trailers is that they reveal way too much of the film, so it's a magnificently welcome shift to see a mainstream Hollywood trailer that is more concerned with mood and atmosphere than a simple retelling of plot.

Another point of interest for the researchers was how Watson chose to focus on moments in the film that were never considered by human creators for earlier trailers. This allowed the filmmakers an entirely novel perspective on their film and a chance to highlight scenes that may not have made it into a more traditional film trailer. As a contrast to Watson's trailer for Morgan posted above, see below for the more conventional human-made trailer.

Morgan | Official Trailer [HD] | 20th Century FOX

But perhaps the most beguiling, and subversive, aspect of Watson's trailer was how much it de-emphasised the monstrous nature of the human/nanotech hybrid. The irony of this entire project is that we have a film where a form of AI turns violent and kills humans, but the AI tasked with making the film's trailer ends up playing down that entire facet of the narrative.

Aside from being a fun experiment in computer-generated creativity, this project also proposes a speedy alternative to a generally costly and time-consuming process. The construction of a film trailer is usually an intensive practice taking several weeks to produce, but this trailer took only 24 hours to construct, from Watson "watching" the film to a human editor delivering the final product.

Making a good film trailer is a delicate balance between art and commerce. If anything this experiment still goes to show that a strong human hand is necessary even when producing what many would determine to be a disposable advertisement. Still, I wouldn't mind getting Watson's perspective on a few sci-fi films that vilify artificial intelligence. Maybe there is a Terminator trailer on the cards that sympathizes with Skynet or a view on 2001: A Space Odyssey where HAL 9000 is the film's hero?

Source: IBM via Popular Science

Don Duncan
Morgan review: A corp? (CIA?) spend years, over two dozen lives, billions, reinventing a SEAL/Special Forces creature. Why? The researchers go out of their way to provoke it to anger. Why? Most of the scientists don't seem to know what the goal of the project is, what they have created. Why? TPTB succeed but keep trying to do it again. Why? Those that know what they have done don't see the danger. Why? None of these questions are answered, or asked.
" ...last year's impressive Ex Machina"
Still have no idea why everyone raves about this highly average film.